When testing the moderation, you are comparing the effects of A on B based on the level of C where A is an exogenous Construct, B is an endogenous construct and C is a moderator. The minimum level of C is two, meaning you will have low level and high level to run separately. For the parametric statistical analysis to be reliable, the minimum sample is 100. So if you have two levels, the minimum sample will be 200.

Thank you a lot !

Is there any publications about that I could mention ?

I'm not sure where the idea of a minimum sample size of 100 per category comes from. The more common recommendation is to have at least 10 observations per parameter in your model (i.e., both the loadings in your measurement model and the coefficients in the structural portion of your model).

I don't know about the source of 100 sample size, either. David seems to be assuming SEM modeling. Because Jean says "simple moderation" I assume that he will run a moderated multiple regression. For MMR, the required sample size is much bigger than you might imagine. For references, Herman Aquinis has published several articles related to this issue.

If you just want to find out the sample size required, you can do that using G*Power. What you are looking for is under linear multiple regression, R-square increase. There are other parameters you should specify, such as expected effect size, the desired power level, and the number of predictors.